fix vits dygraph to static

paddlespeech/t2s/exps/syn_utils.py

@@ -446,6 +446,14 @@ def am_to_static(am_inference,
         am_inference = jit.to_static(
             am_inference, input_spec=[InputSpec([-1], dtype=paddle.int64)])
     elif am_name == 'vits':
+        if am_dataset in {"aishell3", "vctk"} and speaker_dict is not None:
+            am_inference = jit.to_static(
+                am_inference,
+                input_spec=[
+                    InputSpec([-1], dtype=paddle.int64),
+                    InputSpec([1], dtype=paddle.int64),
+                ])
+        else:
             am_inference = jit.to_static(
                 am_inference, input_spec=[InputSpec([-1], dtype=paddle.int64)])
     jit.save(am_inference, os.path.join(inference_dir, am))

paddlespeech/t2s/exps/vits/synthesize_e2e.py

@@ -42,6 +42,9 @@ def evaluate(args):
 
     # frontend
     frontend = get_frontend(lang=args.lang, phones_dict=args.phones_dict)
+    # acoustic model
+    am_name = args.am[:args.am.rindex('_')]
+    am_dataset = args.am[args.am.rindex('_') + 1:]
 
     spk_num = None
     if args.speaker_dict is not None:
@@ -105,9 +108,11 @@ def evaluate(args):
                 for i in range(len(phone_ids)):
                     part_phone_ids = phone_ids[i]
                     spk_id = None
-                    if spk_num is not None:
+                    if am_dataset in {"aishell3", "vctk"
+                                      } and spk_num is not None:
                         spk_id = paddle.to_tensor(args.spk_id)
-                    # wav = vits_inference(text=part_phone_ids, sids=spk_id)
+                        wav = vits_inference(part_phone_ids, spk_id)
+                    else:
                         wav = vits_inference(part_phone_ids)
                     if flags == 0:
                         wav_all = wav